The analysis of upper-body acceleration is a promising and simple technique to quantitatively assess dynamic gait stability. However, this method has rarely been used for people with stroke, probably because of some technical issues still not addressed. We evaluated the root-mean-square (RMS) and harmonic ratio of trunk accelerations for a group of 15 inpatients with subacute stroke who were able to walk (61.4 +/– 14.9 yr) and compared them with those of an age-matched group of nondisabled subjects (65.1 +/– 8.8 yr) and those of a highly functional group of young nondisabled subjects (29.0 +/– 5.0 yr). Small (<2%) but significant (p < 0.03) differences were found in RMS values obtained by applying the two most common computational approaches: (1) averaging among individual-stride RMS values and (2) computing the RMS value over the entire walking trial without stride partitioning. We found that the intersubject dependency of acceleration RMS values by selected walking speed was specific for each group and for each of the three body axes. The analysis of ratios between these three accelerations provided informative outcomes correlated with clinical scores and not affected by walking speed. Our findings are an important step toward transferring accelerometry from human movement analysis laboratories to clinical settings.